The Data Warehouse Toolkit. The Definitive Guide to Dimensional Modeling (Third Edition)

The Data Warehouse Toolkit. The Definitive Guide to Dimensional Modeling (Third Edition)


Ralph Kimball

Margy Ross

Why I Recommend It: 

This is an old book, dating back to the original edition in 1996.  (Ancient in IT land.)  However, this quote alone is worth keeping the book around. 

“We strongly discourage the independent data mart approach. However, often these independent data marts have embraced dimensional modeling because they're interested in delivering data that's easy for the business to understand and highly responsive to queries. So our concepts of dimensional modeling are often applied in this architecture, despite the complete disregard for some of our core tenets, such as focusing on atomic details, building by business process instead of department, and leveraging conformed dimensions for enterprise consistency and integration.”

In other words, independent data marts equal silos.  That is bad.

Topics Covered: 

Ordinarily, I am uncomfortable questioning Kimball and company. That said, I would not universally agree with this statement about geographic location attributes.

“Standardizing the attributes associated with points in space is valuable. However, this is a back room ETL task; you don’t need to unveil the single resultant table containing all the addresses the organization interacts with to the business users. Geographic information is naturally handled as attributes within multiple dimensions, not as a standalone location dimension or outrigger. There is typically little overlap between the geographic locations embedded in various dimensions. You would pay a performance price for consolidating all the disparate addresses into a single dimension.” (Chapter 11, Telecommunications, Geographic Location Dimension.)

Certain telecommunications companies are facilities based, meaning that their customer/prospect base is constrained to businesses located in specific locations (i.e., addresses or buildings). Therefore, it is sometimes advisable to construct a dimension of “serviceable” locations.